Stochastic Optimization for Isolated Microgrid Energy System Design and Control

Abstract

Isolated microgrids are becoming an increasingly popular source of off-grid energy. However, comprehensive modeling frameworks and optimization strategies for finding cost effective design configurations and control policies are lacking in both the cur- rent academic literature and leading microgrid modeling software packages. In this paper, we present a Markov decision process (MDP) based modeling framework for isolated microgrid design and control. Additionally, we propose a methodology for optimally sizing microgrid components, and for finding microgrid control policies that minimize operational costs. For the microgrid design problem, we apply a Thompson Sampling based belief model for rapidly evaluating the performance of a range of microgrid configurations, and for the microgrid control problem, we explore two types of value function approximation (VFA) based policies taken from the field of approx- imate dynamic programing. Simulation results indicate that these strategies can be effective tools for microgrid managers, and can capture realistic microgrid operating behavior. Furthermore, we find that VFA policies can lead to a dramatic reduction in required computation time compared to full MDP models with minimal sacrifices in performance.